The open-source substrate for AI organizations that run themselves — accountably. You set the mission and the guardrails. Your agents do the work.
Autonomy is earned, not assumed.
Most “autonomous company” demos hand a swarm of agents the keys — and hope for the best.
OpenMAO takes the opposite bet.
Every action an agent takes is owned, governed, and auditable. The system widens what agents can do only on a track record it can prove. Accountability isn't the brake on autonomy — it's the road to it.
OpenMAO isn't a pile of features. It's one loop — and each turn, proven in the audit trail, earns the next notch of autonomy.
Every action bounded, owned, non-bypassable.
One agent's lesson becomes the organization's.
Outcomes turn into better decisions.
The org improves its own roles, policies, workflows.
The audited record widens the dial a notch. Always reversible.
Anyone can ship one stage. Almost no one will build the whole loop. That loop is the hard, valuable thing — and the reason OpenMAO exists.
Your agents, frameworks, and tools are workers on top. OpenMAO is the system of record. Anything that can send, spend, deploy, write, or change something routes back through OpenMAO first.
The agent can't reach the side effect without passing the gate. Swap your framework tomorrow — the organization, its memory, and its audit trail are untouched.
OpenMAO starts tight and widens only on earned, audited trust.
OpenMAO is an early local release. Governance, memory, and audit are real and tested. The hard part — genuine self-correction, where an organization diagnoses its own failures and revises its own structure — is the open frontier: specced, seamed into the architecture, and not yet built.
If that problem pulls at you, this is a good place to be.
Build audit, enforcement, and reversibility before widening the dial.
Real side effects are governed only when the perimeter is non-bypassable.
Frameworks and models are interchangeable workers under OpenMAO authority.
Memory, structure, and audit history stay portable and self-hostable.
Self-correction is the open problem. We treat it as the research frontier, not a switch to flip.
The organization you build belongs to you. Not a vendor. Not a cloud.